Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
CS211: Protocol and Systems Design for Wireless and Mobile Networks Instructor: Songwu Lu [email protected] Office: 4531D BH Lectures: 2:00-3:50am M&W office hours: 4:00-5:00pm M&W CS211/Fall 2003 9/29 What this course is about... • Introduce – Internet design philosophy – Wireless networking protocols – Mobile computing system software design – Trendy topics • System programming skills • How to start research CS211/Fall 2003 9/29 A picture of the course coverage Networking fundamentals: Internet philosophy and principles Wireless Protocols -MAC protocol -802.11 Standard - Scheduling - Mobility management, adhoc routing - wireless TCP Mobile Computing - middleware, OS, file sys. - services, applications CS211/Fall 2003 Topical Studies -Wireless security -Sensor networks -QoS and Energy-efficient design -Mesh Networks -MIMO Systems 9/29 Emerging Wireless Networks Internet Backbone Base Station Fixed Host Wireless Cell CS211/Fall 2003 Mobile Host 9/29 Growth of Wireless Users Wireless Phone Subscribers (in millions) 70 60 50 40 30 20 10 0 1991 1993 1995 1997 Source: cellular telecom. Indus. Assn. Wireless Data Subscriber (in millions) 12 10 8 6 4 2 0 CS211/Fall 2003 1997 1998 1999 2000 2001 2002 Source: Strategis Market Res. 9/29 The Wi-Fi Space • It is one of the fastest growing industry sectors – 100,000 public hotspots by 2005 • Most notebooks will have embedded wifi card • Go and check the local hotspots online – www.ezgoal.com/hotspots/ CS211/Fall 2003 9/29 Protocol Stack Application Layer Middleware and OS • Wireless Web, Location Services, etc. Content adaptation, Consistency, File system Transport Layer Wireless TCP Network Layer Mobility, Routing QoS Link Layer & Below CS211/Fall 2003 o Scheduling o MAC 9/29 The Course Description • No required textbook for this course, only a set of papers • Read and discuss – your class participation counts • practice what you have learned – get your hand dirty: do a term project – make your contributions • Heavy workload expected – You are expected to be prepared for each lecture by reading the paper BEFORE coming to the lecture CS211/Fall 2003 9/29 Prerequisites • basic knowledge of packet switched networks & familiarity with TCP/IP protocol suite • adequate programming experience – familiar with C/C++/UNIX – useful reference books: • “Internetworking with TCP/IP, Vol’s I, II, III” by Doug Comer • “TCP/IP Illustrated, Vol’s 1 & 2” by Stevens CS211/Fall 2003 9/29 Course Workload • One midterm, no final exam – Midterm: November 10th, in class. • reading assignment: – 1~2-page summary for the assigned reading of each lecture – 3 strong points, 3 weak points, suggestions – Similar to the paper review process you are going to do for your field in the future • all assignments due 12:00pm(noon) before lecture on the due date – email to [email protected] with subject “cs211: homework #” CS211/Fall 2003 9/29 Course Project • A few big projects – Several topics within each big project to be distributed this Wednesday – 2-3 persons on each topic • Pick a topic and a team by next Monday • Proposal + Checkpoint + Presentation + Final Report CS211/Fall 2003 9/29 Why such projects? • Interact closely within your topic team • Discuss every three weeks within your big project to have the big picture in mind • Stimulate discussions across teams • Most topics are well defined, and you have a good starting point CS211/Fall 2003 9/29 Grading Policy Grading breakdown: • in-class presentation: 10% – 5~10 min each person – Will get an assigned paper (expanding the topic scope of the paper discussed in class) from me • midterm exam: 30% • homework assignments: 20% – There would be 19 assignments, you are expected to turn in at least 15 – The 15 critiques with highest scores to be counted • term project: 40% – proposal 5%, checkpoint 10%, final report 15%, presentation & demo 10% CS211/Fall 2003 9/29 Course policies • Homeworks, project proposals & reports all due 12:00pm on the due date • No late turn-in accepted for credit!!! • No makeup exam!!! Course homepage: http://www.cs.ucla.edu/classes/fall03/cs211/ [email protected] CS211/Fall 2003 9/29 Tips on Doing Research in Graduate School 1. 2. 3. How to do productive research in graduate school What are the bad practices you should avoid Your feedback? CS211/Fall 2003 9/29 The content of this presentation • We take slides and points from many outstanding researchers: Dave Patterson, Richard Hamming, Craig Patridge, Nitin Vaidya, and the many references and sources cited there. They deserve all the credits • I also share some of my own experiences • We need your input and feedback too CS211/Fall 2003 9/29 Caveats • Only opinions from some people. Others may not agree, including your advisors. • Use advice at your own risk • I do not necessarily follow the advice all the time • This presentation may not follow some rules it talks about CS211/Fall 2003 9/29 What is Research, Anyway? • Research is not really about coming up with a nice solution to a hard (possibly new) problem, to show how smart you are. • It is a process: – identifying a research problem – Coming up with a nice/new result (including simulating, implementing, testing your solution) – Writing your results well – Presenting your results – Marketing your work – Engineering is not science, it is about different tradeoff (whether u can do things easier, efficient, more convenient, … at acceptable cost/complexity), precisely true/false is not the main concern CS211/Fall 2003 9/29 A Few EQ Rules • Motivation: “you are indeed interested in PhD research” – Think carefully about your career goal when you start your PhD – NOT: “My family asks me to get a PhD…”, “It is hard to find a job with a MS degree now…”, “I want to hang around in school a little longer…” – We can get you interested in something for some time, but not all the time • Good start: “well begun, half done” – Work harder during the first two years to settle down in research – Have a taste of what is good research; not poisoned by the bad taste – Believe yourself: your mindset has not be “framed” by conventional approaches yet; you can be innovative since you do not know much – You have more energy and can have less distraction at this time • Take the initiative: “you do care about what you are working on” – Do not be afraid to talk to your advisor or others, and let people know the negative results/setbacks etc. » “If u do not talk to these folks, who can u talk to???” » disconnected communication causes more confusion among people – Be honest to research and yourself; do not hide the nasty findings. If you do not understand something, ask; then you will know it. – The reality of “capture effect”: Each advisor has more students than (s)he can handle; whoever is more aggressive gets more feedback more output – Push for the project schedule from your side: call for meetings, set deadlines for internal drafts, look for places where to publish, etc. CS211/Fall 2003 9/29 EQ Rules (Cont’d) • Regular life: “manage your time and life properly” – Shift from “deadline scheduling” to “priority scheduling” – Evaluate your progress periodically. No one else will tell you that you are not efficient – Have a “to-do” list on a daily/weekly/monthly basis – Keep your most productive time-slot during a day to yourself » No interruption even by your advisor, full concentration » Even when the deadline comes CS211/Fall 2003 9/29 How to put yourself into the best position? • Keep yourself informed and networked: “know what is going on and talk to people” – Know the literature on the topic you are working on; not let us tell you what to read. A quick rule “10+10” for breadth and depth: ten top systems/network conferences and ten leading groups – People networking: the best way to be a missionary for your work » Conference is a best place to talk to people. “Do not spend most time to polish your slides/talk there!!” » When people contact you for your work, be responsive. “If you do not care about your work, who should care?” » Attend seminars: people present the “meat” and “dark side” of their work in a talk • Balance between quality and quantity: “make your record without controversy” – Target a top conference each year: show your work quality – Try at least a couple of small conferences: show your productivity » Good way to practice writing, independent research, presentation,… » A nice way to go to scenery places for sightseeing, vacations… CS211/Fall 2003 9/29 Selecting a Problem • Solve a real problem that sb. cares about • Follow the industry technology trend and try to stay ahead of it a little – Bad move: even if technology appears to leave you behind, stand by your problem – Bad move: avoid payoffs of less than 20 years • Working on a new problem is always easier – People have worked on some problems, e.g., congestion control, for years. It is debatably harder for you to jump in and make major contributions • Select a topic that you are interested for some extended period of time, not just for a month • Interdisciplinary topics are always better, they can be very fruitful • Running real experiments to discover new problems • For systems topic, start from yourself: what do you need the systems to do for you? CS211/Fall 2003 9/29 Coming up with a solution • Do not rush for a solution simply based on the literature or what others tell you • Understand the problem better, the solution naturally follows • Use common sense – Do not try to simply combine several existing solutions – Explore new approaches: the alternative/opposite first – Ask questions based on your intuition • Keep things simple unless a very good reason not to – Pick innovation points carefully – Best results are obvious in retrospect “Anyone could have thought of that” • Complexity cost is in longer design, construction, test, and debug – Fast changing field + delays => less impressive results – Bad move: best compliment: it is so complicated, I cannot understand the ideas • Best solutions are a combination of simplicity and depth – Keep the solution core simple – Depth is on second-level issues and fixes • A relevant issue: How do I know mine is different from others – READING PAPERS CS211/Fall 2003 9/29 How to read a paper? Know why you want to read the paper • To know what’s going on – title, authors, abstract – Track a few leading groups/researchers in your area, typically less than 10 is enough – Only a few conferences (and journals): sigcomm, mobicom, infocom, sosp, sigmetrics, mobisys, … • Papers in your broad research area – introduction, motivation, solution description, summary, conclusions – sometimes reading more details useful, but not always • Papers that are directly relevant to your work – read entire paper carefully, and several times CS211/Fall 2003 9/29 What to note • Authors and research group – Need to know where to look for a paper on particular topic • Theme of the solution – Should be able to go back to the paper if you need more info • Approach to performance evaluation • Note any shortcomings • Be critical. It is easy to say nice words about a work, it is harder to identify limitations/flaws CS211/Fall 2003 9/29 In the process of a research project • Get Periodic Reviews/Feedbacks with Others –Talk to people and ask what they think –Give a seminar within your group periodically to collect feedback • Explain the results to your friends, see whether they can grasp your problem and your solution –For both technical people and non-technical people • Exchange emails, publish technical reports CS211/Fall 2003 9/29 Evaluate Quantitatively • If you can’t be proven wrong, then you can’t prove you’re right • Report in sufficient detail for others to reproduce results – can’t convince others if they can’t get same results • For better or for worse, benchmarks shape a field • Good ones accelerate progress – good target for development • Bad benchmarks hurt progress – help real users v. help sales? • Before you run real experiments, do an intuitive analysis – Math does not need to be fancy, as long as it proves the point; in fact, best theory starts from scratch, not from some complex theorem you never heard about CS211/Fall 2003 9/29 Marketing • Publishing papers is not equivalent to marketing • Missionary work: “Sermons” first, then they read papers – Selecting problem is key: “Real stuff” • Ideally, more interest as time passes • Change minds with believable results • Dave Patterson’s experience: industry is reluctant to embrace change – Howard Aiken, circa 1950: “The problem in this business isn’t to keep people from stealing your ideas; its making them steal your ideas!” – Need 1 bold company (often not no. 1) to take chance and be successful • RISC with Sun, RAID with (Compaq, EMC, …) – Then rest of industry must follow • Publicize software: when people use your tools, they know your results – think about how ns-2 and its wireless extension evolve CS211/Fall 2003 9/29 How to write a paper Do unto others as you would have them do unto you When you have truly exceptional results – P == NP – Probably doesn’t matter how you write, people will read it anyway Most papers are not that exceptional Good writing makes significant difference Better to say little clearly, than saying too much unclearly CS211/Fall 2003 9/29 Readability a must • If the paper is not readable, author has not given writing sufficient thought • Two kinds of referees – If I cannot understand the paper, it is the writer’s fault – If I cannot understand the paper, I cannot reject it • Don’t take chances. Write the paper well. • Badly written papers typically do not get read CS211/Fall 2003 9/29 Do not irritate the reader • Define notation before use • No one is impressed anymore by Greek symbols • If you use much notation, make it easy to find – summarize most notation in one place Avoid Using Too Many Acronyms AUTMA ?! You may know the acronyms well. Do not assume that the reader does (or cares to) CS211/Fall 2003 9/29 Writing a draft • First read Strunk and White, then follow these steps; 1. 1-page paper outline, with tentative page budget/section 2. Paragraph map » 1 topic phrase/sentence per paragraph, handdrawn » figures w. captions 3. (Re)Write draft » Long captions/figure can contain details ~ Scientific American » Uses Tables to contain facts that make dreary prose 4. Read aloud, spell check & grammar check (MS Word; Under Tools, select Grammar, select Options, select “technical” for writing style vs. “standard”; select Settings and select) 5. Get feedback from friends and critics on draft; go to 3. • www.cs.berkeley.edu/~pattrsn/talks/writingtips.html CS211/Fall 2003 9/29 How to write a systems paper • Provide sufficient information to allow people to reproduce your results – people may want to reproduce exciting results – do not assume this won’t happen to your paper – besides, referees expect the information • Do not provide wrong information • Sometimes hard to provide all details in available space – may be forced to omit some information – judge what is most essential to the experiments – cite a tech report for more information CS211/Fall 2003 9/29 Discuss related work • Explain how your work relates to state of the art • Discuss relevant past work by other people too • Remember, they may be reviewing your paper. – Avoid: The scheme presented by FOO performs terribly – Prefer: The scheme by FOO does not perform as well in scenario X as it does in scenario Y • Avoid offending people, unless you must CS211/Fall 2003 9/29 Tell them your shortcomings • If your ideas do not work well in some interesting scenarios, tell the reader • People appreciate a balanced presentation CS211/Fall 2003 9/29 How to write weak results • If results are not that great, come up with better ones • Do not hide weak results behind bad writing – Be sure to explain why results are weaker than you expected • If you must publish: write well, but may have to go to secondbest conference – Only a few conferences in any area are worth publishing in – Too many papers in poor conferences bad for your reputation – Just because a conference is “IEEE” or “ACM” or “International” does not mean it is any good • If results not good enough for a decent conference, rethink your problem/solution CS211/Fall 2003 9/29 Miscellaneous • Read some well-written papers – award-winning papers from conferences • Avoid long sentences • If you have nothing to say, say nothing – don’t feel obliged to fill up space with useless text – if you must fill all available space, use more line spacing, greater margins, bigger font, bigger figures, anything but drivel CS211/Fall 2003 9/29 How to present a poster • Answer Five Heilmeier Questions 1. What is the problem you are tackling? 2. What is the current state-of-the-art? 3. What is your key make-a-difference concept or technology? 4. What have you already accomplished? 5. What is your plan for success? • Do opposite of Bad Poster commandments – Poster tries to catch the eye of person walking by • 9 page poster might look like Problem Statement State-ofthe-Art Accomplish Title and -ment # 1 Visual logo Accomplish Plan for -ment # 3 Success CS211/Fall 2003 Key Concept Accomplish -ment # 2 Summary & Conclusion 9/29 How to present a paper (at a conference) Objectives, in decreasing order of importance • Keep people awake and attentive – everything has been tried: play fiddle, cartoons, jokes – in most cases, extreme measures should not be needed – humor can help • Get the problem definition across – people in audience may not be working on your problem Explain your general approach most productive use of your time Dirty details most people in the audience probably do not care a typical conference includes 30+ paper presentations, yours could be the N-th CS211/Fall 2003 9/29 How many slides? • Depends on personal style • Rules of thumb – 1 slides for 1-2 minutes – Know your pace • I tend to make more slides than I might need, and skip the not-so-important ones dynamically • Anticipate technical questions, and prepare explanatory slides CS211/Fall 2003 9/29 How to present a paper • Practice makes perfect (or tolerable) • May need several trials to fit your talk to available time » particularly if you are not an experienced speaker English issue Accent may not be easy to understand Talk slowly Easier said than done I have a tough time slowing down myself CS211/Fall 2003 9/29 The rest of the notes Overview/Review: • Internet protocol stack • IP protocol • TCP protocol If you forget these materials, go back and review what you learned in CS118 ASAP CS211/Fall 2003 9/29 Packet Switched Networks • Hosts send data in packets • network supports all data communication services by delivering packets Host – Web, email, multimedia video Host Application Host Web Host Host email CS211/Fall 2003 9/29 One network application example [email protected] [email protected] msg CS211/Fall 2003 9/29 What is happening inside ? [email protected] msg email [email protected] Transport protocol Transport protocol Network protocol Physical net CS211/Fall 2003 Network protocol Network protocol Physical net Network protocol physical net 9/29 Layered Network Architecture • network consists of geographically distributed hosts and switches (nodes) • Nodes communicate with each other by standard protocols A host A C switch B C D network topology CS211/Fall 2003 B physical connectivity Protocol layers 9/29 a picture of protocol layers A Application (data) header data Transport segment header DATA network packet DATA header Ethernet frame tail B physical connectivity What’s in the header: info needed for the protocol’s function CS211/Fall 2003 9/29 TCP/IP Protocol Suite • IP Protocol: Inter-networking protocol – RFC791 • TCP Protocol: reliable transport protocol – RFC793 CS211/Fall 2003 9/29 Why IP • a number of different network technologies developed in early 70’s: ARPAnet, Ethernet, Satnet, PRnet • different trans. media: copper, radio, satellite • different protocol designs, e.g. • ARPAnet: reliable message delivery • Ethernet: unreliable packet delivery • under different administrative control CS211/Fall 2003 9/29 Fundamental Goal of IP • developing an effective technique for multiplexed utilization of all existing networks – no centralized control – no changes to individual subnets To read next time “The Design Philosophy of Internet Protocols” by Dave Clark, SIGCOMM'88 CS211/Fall 2003 9/29 The picture of the world according to IP application protocols transport layer protocols TCP UDP universal datagram delivery transport (end-to-end) IP inter-network layer subnets hardware-specific network technologies ethernet token-ring FDDI dialup ATM CS211/Fall 2003 9/29 IP Packet Header Format Type-of-Service identifier time-to-live total length MF IHL DF vers. protocol fragment offset checksum source address destination address options (variable length) padding data CS211/Fall 2003 9/29 IP: two basic functions • a globally unique address for each reachable interface • datagram delivery from any host to any other host(s) two supporting protocols • ICMP (Internet Control Message Protocol) • ARP (Address Resolution Protocol) CS211/Fall 2003 9/29 Fundamental challenge: How to scale better • Original design: two levels of hierarchy, network, host • Observed problems: – class-based address assignment infeasible – too many networks visible at the top level • two approaches: subnetting & (CIDR) supernetting CS211/Fall 2003 9/29 Longer-Term Scaling issues • We've run out of all IPv4 unicast space – far before theoretical limit of 4 billion hosts, due to inevitable inefficiency of address block allocation • Short term patch: NAT boxes • One long term solution: IP version 6 – expanded addressing capability: 16 bytes – cleanup of IPv4 design after 15 years of running experience – improved support for options/extensions CS211/Fall 2003 9/29 The IPv6 Header Version Priority Payload Length Flow Label Next Header Hop Limit Source Address Destination Address 32 bits CS211/Fall 2003 9/29 The IPv4 Header Version Hdr Len Prec TOS Total Length Identification Flags Fragment Offset Time to Live Protocol Header Checksum Source Address Destination Address Options Padding 32 bits shaded fields are absent from IPv6 header CS211/Fall 2003 9/29 TCP: Transmission Control Protocol • a transport protocol – IP delivers packets “from door to door” – TCP provides full-duplex, reliable byte-stream delivery between two application processes Application process More terminology: • TCP segment • Max. segment size (MSS) Write bytes Read bytes TCP TCP Send buffer Receive buffer segment CS211/Fall 2003 Application process segment 9/29 TCP: major functionalities • Header format • Connection Management • Open, close • State management • Reliability management • Flow and Congestion control • Flow control: Do not flood the receiver’s buffer • Congestion control: Do not stress the network by sending too much too fast CS211/Fall 2003 9/29 TCP header format 0 3 1 1 6 IP header source port destination port Data sequence number acknowledgment number Hlen unused u a p r s f r c s s y i g k h t n n checksum window size urgent pointer Options (viable length) data CS211/Fall 2003 9/29 opening a connection: three-way hand-shake client open request(x) server Passive open ack(x+1) + request(y) ack(y+1) (now in estab. state) CS211/Fall 2003 enter estab. state 9/29 Closing a TCP Connection I-finished(M) ACK (M+1) I-finished(N) ack(N+1) wait for 2MSL before deleting the conn state CS211/Fall 2003 Done, delete conn. state 9/29 Mechanisms for Reliability Management • • • • Sequence number Acknowledgment number Error detection at the receiver side Retransmission timeout CS211/Fall 2003 9/29 TCP Flow and Congestion Control • Window-based protocol • Flow control is easy: set the sender’s window no larger than the advertised window by the receiver • 4 algorithms in TCP congestion control – Control congestion window variable: cwnd – slow start, congestion avoidance, fast retransmit and fast recovery, retransmission upon timeout • Sender_window=min(adv_win, cwnd) CS211/Fall 2003 9/29 Slow Start & Congestion Avoidance • start conservatively: cwnd <= min(2*SMSS bytes, 2 segments) • when cwnd <ssthresh, use slow start: – increase cwnd exponentially to quickly fill up the pipe: upon receiving each ACK, cwnd+=SMSS; • when cwnd > ssthresh, use congestion avoidance – cwnd += SMSS*SMSS/cwnd; – continue until loss is detected CS211/Fall 2003 9/29 Fast Retransmit • When the 3rd DUP_ACK is received, ssthresh=max(FlightSize/2, 2*SMSS) • ReXmit the lost segment, set cwnd=ssthresh+3*SMSS • Design questions: • why FlightSize, not cwnd ? –FlightSize: data sent but not yet acked • Why add 3 SMSS to cwnd ? CS211/Fall 2003 9/29 Fast Recovery • For each additional DUP_ACK: – cwnd+=SMSS; (why ?) – transmit a new segment if min(cwnd, rwnd) permits • When a NEW ACK arrives, – cwnd=ssthresh; (why ?) CS211/Fall 2003 9/29 Retransmission Timeout • Initial design: – RTT=*old_RTT+ (1-)*New_RTT_sample – RTO= *RTT; = 2 for original spec – variation in RTT: ~1/(1-L); factor 4, for L=50%; factor 10, for L=80%; load <= 30% for =2. • RTO improvement – in addition to mean, also estimate the deviation of RTT • Diff=New_RTT_sample - old-RTT; • Smoothed_RTT=old_RTT+1/8*Diff • Dev=old_RTT+1/4*(|Diff|-Old_Dev) – RTO = Smoothed_RTT+4*Dev CS211/Fall 2003 9/29 Karn’s Algorithm • how to measure RTT in retransmission cases? – take the delay between the first (last) transmission and final ack? – do not update SRTT in case of retransmission? • Karn’s algorithm: – do not take RTT samples in case of retransmission – double the retransmission timer for next packet, till one can get a RTT sample without retransmission CS211/Fall 2003 9/29 Putting all together: RFC2581 • how TCP congestion control works – Start with slow start for bootstrapping phase: quickly open up the window – At ssthresh, switch to congestion avoidance – When 3rd duplicate ACK is received (indicating a packet loss), use fast retransmit; if more than 3 duplicate ACKs, use fast recovery – Upon retransmission timeout (indicating a packet loss too): cwnd=1, binary exponential backoff CS211/Fall 2003 9/29